A method for tracking weld seams in structural components based on machine vision sensors

Abstract

This invention provides a method for tracking weld seams in structural components based on machine vision sensors, comprising the following steps: optimizing the weld joint form by improving the blunt edge bevel to a bevel without a blunt edge; establishing a real-time tracking and dynamic recognition system for weld seam features based on machine vision sensors to identify assembly gaps, blunt edge dimensions, bevel angles, and welding trajectories; establishing a welding parameter library for different assembly gaps according to blunt edge dimensions; acquiring images of the welding area of ​​the structural component collected by machine vision sensors before and during welding, and applying the real-time tracking and dynamic recognition system for weld seam features to acquire assembly gaps, blunt edge dimensions, bevel angles, and weld seam trajectories; automatically matching the parameter library based on the recognition results to obtain welding parameters and sending them to the welding torch control unit to control the welding robot to weld the structural component; simultaneously, calculating the weld seam offset based on a linear laser sensor and the weld seam trajectory, and sending the weld seam offset to the welding torch control unit for welding torch position compensation.

Description

Technical Field

[0001] This invention relates to the field of welding technology for hydraulic support structural components in underground coal mines, and more specifically, to a method for tracking weld seams in structural components based on machine vision sensors. Background Technology

[0002] As core equipment in mine support systems, the welding quality of hydraulic support structural components directly determines the safety and service life of the equipment. These components are mostly made of high-strength steel thick plates, with complex weld seam configurations and large fluctuations in assembly gaps, requiring extremely high welding precision. Traditional welding operations often use weld seam tracking systems based on electric arcs and analog sensors. However, these systems cannot establish predictive models of welding thermal cycles and changes in assembly gaps, making it difficult to predict the impact of welding parameters on penetration quality. They also cannot achieve high-precision real-time identification of assembly gaps, blunt edge dimensions, and welding trajectories, and their ability to perceive weld seam features is insufficient under complex working conditions.

[0003] With the development of machine vision, numerous solutions for real-time tracking of welding bevels and welds based on machine vision have been proposed. Examples include CN113510412B, a detection system, method, and storage medium for identifying weld conditions; and CN116228858A, a method for automatic bevel detection and welding based on machine vision. However, these solutions all acquire bevel parameters before welding and calculate welding parameters based on these parameters. They cannot dynamically adjust parameters according to structural features such as plate thickness, assembly gap, and blunt edge dimensions. This results in problems such as low efficiency in multi-layer, multi-pass welding, shallow penetration, and low weld penetration pass rate, making it difficult to meet the high-precision welding requirements of hydraulic support structural components.

[0004] Meanwhile, traditional welded joints are mostly designed with a 2mm blunt bevel, but conventional linear laser vision sensors have a blind spot for detecting the blunt edge of the bevel, which can easily lead to deviations in the detection of the bevel's geometric dimensions. Therefore, there is an urgent need to develop a machine vision weld tracking technology adapted to the welding conditions of hydraulic support structural components to solve the above-mentioned technical pain points.

[0005] In order to solve the above problems, people have been seeking an ideal technological solution. Summary of the Invention

[0006] Therefore, it is necessary to provide a structural weld seam tracking method based on machine vision sensors to address the above-mentioned technical problems, so as to achieve high-precision real-time perception of weld seam features, accurate compensation of welding trajectory and dynamic matching of welding parameters, improve the welding quality and efficiency of hydraulic support structural components and break through the bottleneck of traditional welding technology.

[0007] To achieve the above objectives, a first aspect of the present invention provides a method for tracking weld seams in structural components based on machine vision sensors, comprising the following steps: The welding joint design was optimized by replacing the blunt edge bevel with a non-blunt edge bevel to eliminate blind spots in the machine vision sensor detection. A real-time tracking and dynamic recognition system for weld features based on machine vision sensors was built to identify assembly gaps, blunt edge dimensions, bevel angles, and welding trajectories. Establish a welding parameter library based on the blunt edge size for different assembly gaps, different plate thicknesses, and different bevel angles; Before and during welding, images of the welding area of ​​the structural components are acquired by machine vision sensors. A real-time tracking and dynamic recognition system for weld features is applied to obtain the assembly gap, blunt edge size, bevel angle, and weld trajectory. The plate thickness is obtained, and based on the plate thickness, assembly gap, blunt edge size, and bevel angle as structural features, the parameters are automatically matched to obtain welding parameters, which are then sent to the welding torch control unit to control the welding robot to weld the structural components. At the same time, the weld offset is calculated based on the weld trajectory, and the weld offset is sent to the welding torch control unit for welding torch position compensation.

[0008] Furthermore, a real-time tracking and dynamic recognition system for weld features based on machine vision sensors was established to identify assembly gaps, blunt edge dimensions, and welding trajectories, including: A target detection model for recognizing assembly gaps, blunt edge dimensions, and welding contours is constructed. The target detection model is based on the YOLO neural network. In addition to the existing classification and bounding box regression branches of the detection head, gap regression, blunt edge dimension regression, and trajectory branches are added. The gap branch is used to output the assembly gap; the blunt edge dimension regression branch is used to output the blunt edge dimension; and the trajectory branch is used to output the weld trajectory points. Furthermore, in each branch of the detection head, asymmetric strip convolution along the laser direction is introduced to replace the standard convolution.

[0009] The beneficial effects of this invention are as follows: This invention changes the weld joint from a blunt edge to one without a blunt edge, eliminating the sensor detection blind zone, significantly improving the accuracy of bevel geometry detection, and providing accurate data support for subsequent welding parameter adjustments.

[0010] Welding parameter libraries were established based on the blunt edge size for different assembly gaps, plate thicknesses, and bevel angles, enabling millisecond-level dynamic adjustment of welding parameters. The parameter library covers different blunt edge and assembly gap conditions, and has strong adaptability.

[0011] A real-time tracking and dynamic recognition system for weld features based on machine vision sensors was built to identify assembly gaps, blunt edge dimensions, bevel angles, and welding trajectories. Based on the assembly gaps, blunt edge dimensions, and bevel angles, a welding parameter library was searched to achieve automatic matching of welding parameters. This system breaks through the technical bottlenecks of low efficiency and shallow penetration in traditional multi-layer and multi-pass welding, and significantly improves welding operation efficiency.

[0012] Furthermore, based on the YOLO neural network model, in addition to the existing classification and bounding box regression branches of the detection head, gap regression, blunt edge size regression, and trajectory branches are added. The gap branch is used to output the assembly gap; the blunt edge size regression branch is used to output the blunt edge size; and the trajectory branch is used to output the weld trajectory points. In each branch of the detection head, asymmetric strip convolution along the laser direction is introduced to replace the standard convolution, thereby achieving high-precision recognition of assembly gap, blunt edge size, and welding trajectory. This solves the problem that traditional sensor systems cannot perceive the characteristics of welds under complex working conditions, and achieves high recognition accuracy and strong real-time performance. Attached Figure Description

[0013] Figure 1 This is a flowchart of a structural weld seam tracking method based on a machine vision sensor, as described in Example 1. Figure 2 This is a schematic diagram comparing the traditional 2mm blunt edge bevel with the improved blunt edge bevel. Figure 3 Establish a logic block diagram for hierarchical classification of the welding parameter library; Figure 4 This is a schematic diagram of the experimental platform for welding high-strength steel thick plates for hydraulic supports. Detailed Implementation

[0014] The technical solution of the present invention will be further described in detail below through specific embodiments. Example 1

[0015] This embodiment provides a method for tracking weld seams in structural components based on machine vision sensors, such as... Figure 1 As shown, it includes the following steps: Step S0: Optimize the welded joint form by changing the blunt edge bevel to a bevel without a blunt edge, thereby eliminating the blind zone of the machine vision sensor and improving the accuracy of bevel geometry detection.

[0016] Specifically, such as Figure 2 As shown, taking the traditional 2mm blunt edge bevel as an example, while keeping the bevel angle unchanged, the traditional single-sided V-shaped bevel with blunt edge is improved into a single-sided V-shaped bevel without blunt edge. The bottom of the bevel adopts a rounded transition, the bevel angle deviation is ≤±1°, and the surface roughness Ra≤12.5μm.

[0017] Step S1: Build a real-time tracking and dynamic recognition system for weld features based on machine vision sensors to identify assembly gaps, blunt edge dimensions, bevel angles, and welding trajectories.

[0018] In practical implementation, a real-time tracking and dynamic recognition system for weld seam features based on machine vision sensors is built, including: A target detection model for recognizing assembly gaps, blunt edge dimensions, and welding contours is constructed. The target detection model is based on the YOLO neural network. In addition to the existing classification and bounding box regression branches of the detection head, gap regression, blunt edge dimension regression, and trajectory branches are added. The gap branch is used to output the assembly gap; the blunt edge dimension regression branch is used to output the blunt edge dimension; and the trajectory branch is used to output the weld trajectory points. Furthermore, in each branch of the detection head, asymmetric strip convolution along the laser direction is introduced to replace the standard convolution.

[0019] It is understandable that introducing an asymmetric strip convolution optimization detection head structure along the laser direction can enhance the model's ability to identify transverse features of the weld, and achieve simultaneous identification and dynamic tracking of gap, blunt edge size and welding trajectory.

[0020] In practical implementation, the machine vision sensor includes a linear laser sensor and an industrial camera. The linear laser sensor is positioned 100-200mm in front of the welding robot. The industrial camera and the linear laser sensor sample synchronously at a sampling frequency ≥50Hz. It can be understood that the image acquired by the industrial camera is a laser stripe image, which contains complete cross-sectional information such as the bevel edge, gap, blunt edge, and bevel angle. By extracting features and performing reverse modeling on the stripe image, the actual shape of the bevel at that cross-section can be reconstructed, thus achieving real-time monitoring of the bevel geometry. This allows the welding robot to adjust its parameters promptly based on the monitoring results.

[0021] Step S2: Establish a welding parameter library for different assembly gaps, different plate thicknesses, and different bevel angles according to the blunt edge size.

[0022] In one embodiment, the welding parameter library includes welding current, welding voltage, oscillation mode, wire feed speed, welding speed, and corresponding heat input.

[0023] In one embodiment, such as Figure 3 As shown, the parameter library is established according to the following hierarchical method: Establish a welding parameter library for different assembly gaps, different plate thicknesses, and different bevel angles when the blunt edge is 0mm. Based on the parameter library with a blunt edge of 0mm, a welding parameter library is established for different assembly gaps, different plate thicknesses, and different bevel angles when the blunt edge is 1mm. Based on the parameter library with a blunt edge of 1mm, a welding parameter library is established for different assembly gaps, different plate thicknesses, and different bevel angles when the blunt edge is 2mm.

[0024] Similarly, a welding parameter library is established based on the blunt edge size for different assembly gaps, different plate thicknesses, and different bevel angles.

[0025] Understandably, the heat input (KJ / mm) corresponding to each parameter can be calculated based on the welding process requirements and the material of the hydraulic support structural components, thereby improving the parameter library and adapting to the welding needs of high-strength steels such as Q550D and Q690D.

[0026] Step S3: Before welding, acquire images of the welding area of ​​the structural component collected by the machine vision sensor, apply the real-time tracking and dynamic recognition system for weld features to acquire the assembly gap, blunt edge size, bevel angle and weld trajectory; acquire the plate thickness, and based on the plate thickness, assembly gap, blunt edge size and bevel angle as structural features, perform automatic matching of the parameter library to obtain welding parameters and send them to the welding gun control unit to control the welding robot to weld the structural component.

[0027] It is understandable that by using a real-time tracking and dynamic recognition system for weld features, the system can identify assembly gaps, blunt edge dimensions, bevel angles, and welding trajectories; and by searching the welding parameter library based on assembly gaps, blunt edge dimensions, and bevel angles, it can achieve automatic matching of welding parameters, breaking through the technical bottlenecks of low efficiency and shallow penetration in traditional multi-layer and multi-pass welding, and significantly improving welding operation efficiency.

[0028] It should be noted that before welding, the weld trajectory point is 0 in the output of the real-time tracking and dynamic recognition system for weld features.

[0029] Step S4: During the welding process, images of the welding area of ​​the structural component are acquired by machine vision sensors. A real-time tracking and dynamic recognition system for weld features is applied to acquire the assembly gap, blunt edge dimension, bevel angle, and weld trajectory. The plate thickness is acquired, and based on the plate thickness, assembly gap, blunt edge dimension, and bevel angle as structural features, parameter library is automatically matched to obtain welding parameters, which are then sent to the welding torch control unit to control the welding robot to weld the structural component. Simultaneously, the weld offset is calculated based on the weld trajectory, and the weld offset is sent to the welding torch control unit for welding torch position compensation.

[0030] It is understandable that by placing a linear laser sensor in front of the welding head, the sensor can detect the left and right offsets and the high and low offsets of the weld in advance, and transmit the weld direction to the welding torch control unit in real time. The reverse modeling method is used to monitor the geometry of the bevel in real time, and the welding robot is guided in a timely manner to compensate for the workpiece offset based on the monitoring results, so as to complete the welding operation of the complex hydraulic support structure. Example 2

[0031] This embodiment provides an example of the structural weld seam tracking method based on machine vision sensors described in Embodiment 1.

[0032] The specific steps are as follows: The first step addresses the issue of blind spots in linear laser sensors caused by traditional blunt edge bevels by optimizing the welded joint design, as detailed below: Existing defects: Traditional 2mm blunt edge single-sided V-groove (40°) will create a 0.5-1mm blind zone for inspection, with an inspection error of 0.1-0.2mm, which can easily lead to incomplete penetration defects.

[0033] Optimized design: The traditional 2mm blunt edge single-sided V-shaped bevel is improved to a single-sided V-shaped bevel without blunt edge (maintaining a 40° bevel angle), with a bevel angle deviation of ≤±1° and a surface roughness Ra≤12.5μm, completely eliminating the detection blind zone.

[0034] Effect verification: After optimization, the blind spot of the detection is completely eliminated, the detection error of the assembly gap and bevel size is ≤0.05mm, the accuracy is improved by more than 60%, and the incomplete penetration defect is reduced, laying the foundation for subsequent follow-up welding.

[0035] The second step is the construction and debugging of a weld seam tracking system based on a linear laser sensor.

[0036] System Hardware Selection and Installation: A linear laser sensor machine vision real-time tracking and dynamic weld seam feature recognition system is being built. The core hardware selection and installation are as follows to ensure high-precision recognition requirements: Linear laser sensor: The Keyence IV2 series laser displacement sensor is selected, with a laser wavelength of 650nm, a measurement range of 0-50mm, a sampling frequency of 60Hz, a measurement accuracy of ±0.01mm, and a resolution of 0.001mm. It is front-mounted 150mm from the front of the welding torch, at a 45° angle with the welding torch axis to avoid interference from smoke and arc light and ensure that the detection field of view covers the welding bevel area.

[0037] Visual processing unit: It adopts an Intel Core i7-12700H processor paired with an NVIDIA Jetson AGXXavier visual chip, with a response time of ≤5ms, enabling real-time processing of image data.

[0038] Welding robot and control unit: The KUKA KR C5 six-axis welding robot (repeat positioning accuracy ±0.03mm, welding torch fine adjustment accuracy ±0.05mm) is selected, which is compatible with MAG welding; the Siemens S7-1500 PLC is used as the welding torch control unit, and high-speed communication between units is achieved through the EtherCAT protocol (1000Mbps).

[0039] Auxiliary equipment: Equipped with fume purification, arc light shielding devices and ±0.02mm precision positioning fixtures to ensure detection accuracy and workpiece fixation stability.

[0040] System software debugging and parameter setting: The software is developed using the C++ programming language, and a feature recognition algorithm is built based on the OpenCV machine vision library. Parameter optimization and modeling are implemented using Python. Debugging is as follows: Image preprocessing: Adaptive threshold segmentation and Gaussian filtering algorithm are used to remove noise interference. Canny algorithm (threshold 80-200) is used to extract bevel edge features with recognition accuracy ≤0.05mm.

[0041] Tracking and communication debugging: Dynamic tracking of weld seam trajectory is achieved through feature point matching and trajectory fitting (delay ≤ 8ms, offset exceeding 0.1mm triggers compensation command); debugging of communication between each unit ensures no data loss and no delay (response ≤ 10ms), and put into use after successful simulation verification.

[0042] Construct an experimental platform for welding high-strength steel thick plates with hydraulic supports, such as... Figure 4 As shown, the above-mentioned weld seam tracking system based on a linear laser sensor is deployed in the platform.

[0043] The third step: Building a dynamic matching database for assembly gap, blunt edge dimensions, and welding parameters.

[0044] To achieve precise matching between welding parameters and assembly gaps and blunt edge dimensions, a dynamic matching database is constructed, following these steps: Grading standards: Taking into account the slight deviations that may exist after optimization without blunt edges, blunt edges are divided into three levels (0-0.5mm, 0.5-1.0mm, 1.0-2.0mm); the assembly gap is set to 0-6mm (0.5mm interval), which is suitable for Q690 steel plates with a thickness of 12-25mm.

[0045] Parameter acquisition: MAG welding (Ar 80% + CO 20% mixed shielding gas, flow rate 20-25 L / min) was used to collect the optimal welding parameters under different working conditions: Level 1 blunt edge, gap 0-1 mm: current 240-260 A, voltage 24-26 V, speed 30-35 cm / min, heat input 0.79-1.08 KJ / mm; Level 2 blunt edge, gap 2-3 mm: current 200-220 A, voltage 20-22 V, speed 20-24 cm / min, heat input 0.80-1.16 KJ / mm; Level 3 blunt edge, gap 4-6 mm: current 190-210 A, voltage 20-22 V, speed 15-20 cm / min, heat input 0.91-1.48 KJ / mm.

[0046] Database construction: The parameters are stored using an SQL Server database, and a parameter correction module is added (correction accuracy ±1A, ±0.1V). The response time is ≤3ms, and the corresponding welding parameters can be automatically matched based on sensor detection data.

[0047] The fourth step is the practical application of weld seam tracking and dynamic parameter adjustment.

[0048] Based on the system built in the above embodiments, the optimized joints, and the constructed database, the tracking of weld seams and dynamic adjustment of parameters of hydraulic support structural components are realized. The complete process is as follows: 1. Workpiece positioning: The top beam of the 20mm thick hydraulic support is assembled without blunt edges (assembly gap 2.5mm, blunt edge 1.0mm), and fixed by a ±0.02mm precision positioning fixture. Clean the bevel surface of rust, oil and other impurities.

[0049] 2. System Initialization: Start each device to complete initialization, debug the linear laser sensor to detect the field of view, and call the initial parameters (current 210A, voltage 20V, etc.) from the dynamic matching database.

[0050] 3. Trajectory tracking and offset compensation: The sensor acquires images of the welding area in real time, and the vision processing unit accurately identifies features (accuracy ≤ 0.05mm) to guide the welding robot to move along the weld trajectory (tracking accuracy ≤ 0.03mm); after detecting a workpiece offset of 0.15mm, the compensation value is calculated through reverse modeling, and the robot is controlled to fine-tune the welding torch to complete the compensation (compensation accuracy ≤ 0.2mm).

[0051] 4. Parameter Adjustment and Quality Monitoring: Welding parameters are dynamically adjusted based on the test data (response time ≤ 10ms). For example, when the assembly gap increases to 3.5mm, the current is adjusted to 200A and the voltage to 18.5V simultaneously. Parameter fluctuations are monitored in real time, and an alarm is triggered to suspend welding when the parameters exceed the tolerance. After welding is completed, ultrasonic testing is performed. The weld is free of defects, and the tensile strength is ≥700MPa and the impact toughness is ≥50J / cm², meeting the quality requirements.

[0052] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the scope of the technical solutions claimed in the present invention.

Claims

1. A method for tracking weld seams in structural components based on machine vision sensors, characterized in that, Includes the following steps: The welding joint design was optimized by replacing the blunt edge bevel with a non-blunt edge bevel to eliminate blind spots in the machine vision sensor detection. A real-time tracking and dynamic recognition system for weld features based on machine vision sensors was built to identify assembly gaps, blunt edge dimensions, bevel angles, and welding trajectories. Establish a welding parameter library based on the blunt edge size for different assembly gaps, different plate thicknesses, and different bevel angles; Before and during welding, images of the welding area of ​​the structural components are acquired by machine vision sensors. A real-time tracking and dynamic recognition system for weld features is applied to obtain the assembly gap, blunt edge size, bevel angle, and weld trajectory. The plate thickness is obtained, and based on the plate thickness, assembly gap, blunt edge size, and bevel angle as structural features, the parameters are automatically matched to obtain welding parameters, which are then sent to the welding torch control unit to control the welding robot to weld the structural components. At the same time, the weld offset is calculated based on the weld trajectory, and the weld offset is sent to the welding torch control unit for welding torch position compensation.

2. The method for tracking weld seams in structural components based on machine vision sensors according to claim 1, characterized in that, The machine vision sensor includes a linear laser sensor and an industrial camera. The linear laser sensor is positioned 100-200mm in front of the welding robot. The industrial camera and the linear laser sensor sample synchronously, and the sampling frequency is ≥50Hz.

3. The method for tracking weld seams in structural components based on machine vision sensors according to claim 2, characterized in that, The welding parameter library includes welding current, welding voltage, oscillation mode, wire feed speed, welding speed, and corresponding heat input.

4. A method for tracking weld seams in structural components based on machine vision sensors according to any one of claims 1-3, characterized in that, A real-time tracking and dynamic recognition system for weld seam features based on machine vision sensors was established to identify assembly gaps, blunt edge dimensions, and welding trajectories, including: A target detection model for recognizing assembly gaps, blunt edge dimensions, and welding contours is constructed. The target detection model is based on the YOLO neural network. In addition to the existing classification and bounding box regression branches of the detection head, gap regression, blunt edge dimension regression, and trajectory branches are added. The gap branch is used to output the assembly gap; the blunt edge dimension regression branch is used to output the blunt edge dimension; and the trajectory branch is used to output the weld trajectory points. Furthermore, in each branch of the detection head, asymmetric strip convolution along the laser direction is introduced to replace the standard convolution.

5. The method for tracking weld seams in structural components based on machine vision sensors according to claim 1, characterized in that, Optimize the welded joint design by replacing the blunt edge bevel with a non-blunt edge bevel, including: While keeping the bevel angle unchanged, the traditional single-sided V-shaped bevel with blunt edge is improved into a single-sided V-shaped bevel without blunt edge. The bottom of the bevel adopts a rounded transition, the bevel angle deviation is ≤±1°, and the surface roughness Ra≤12.5μm.

Images